Long-term summer temperature variations in the Pyrenees
Two hundred and sixty one newly measured tree-ring width and density series from living and dry-dead conifers from two timberline sites in the Spanish Pyrenees were compiled. Application of the regional curve standardization method for tree-ring detrending allowed the preservation of inter-annual to multi-centennial scale variability. The new density record correlates at 0.53 (0.68 in the higher frequency domain) with May–September maximum temperatures over the 1944–2005 period. Reconstructed warmth in the fourteenth to fifteenth and twentieth century is separated by a prolonged cooling from ∼1450 to 1850. Six of the ten warmest decades fall into the twentieth century, whereas the remaining four are reconstructed for the 1360–1440 interval. Comparison with novel density-based summer temperature reconstructions from the Swiss Alps and northern Sweden indicates decadal to longer-term similarity between the Pyrenees and Alps, but disagreement with northern Sweden. Spatial field correlations with instrumental data support the regional differentiation of the proxy records. While twentieth century warmth is evident in the Alps and Pyrenees, recent temperatures in Scandinavia are relatively cold in comparison to earlier warmth centered around medieval times, ∼1450, and the late eighteenth century. While coldest summers in the Alps and Pyrenees were in-phase with the Maunder and Dalton solar minima, lowest temperatures in Scandinavia occurred later at the onset of the twentieth century. However, fairly cold summers at the end of the fifteenth century, between ∼1600–1700, and ∼1820 were synchronized over Europe, and larger areas of the Northern Hemisphere.
Increasing data paucity back in time is also a general rule at the regional-scale including the European sector (hereinafter EU), for which the pioneering work by Lamb (1965) is still referenced for the occurrence of relatively warm EU conditions during medieval times, the so-called Medieval Warm Period (MWP). Hence, the annual-resolved comparison of the EU 2003 summer heat (Luterbacher et al. 2004; Schär et al. 2004) with earlier warm periods still lacks confidence (Büntgen et al. 2006b). Nevertheless, significant progress has recently been made in understanding EU climate variability over the past 150 to about 500 years through studies of long instrumental station records (Auer et al. 2007 and references therein), documentary evidences (Brázdil et al. 2005 and references therein), tree-ring chronologies (Büntgen et al. 2007a; Frank and Esper 2005b), and multi-proxy compilations (Luterbacher et al. 2004), for example. Detailed knowledge of atmospheric circulation patterns over recent centuries is reported for the Alps (Casty et al. 2005), and the North Atlantic/EU sector (Luterbacher et al. 2002; Raible et al. 2006 and references therein). Reconstructions of synoptic-scale precipitation variability are developed using grid-box data (Pauling et al. 2006), and more locally using tree-ring chronologies (Esper et al. 2007; Wilson et al. 2005). In southern EU, temperature sensitive proxies with tree-rings in particular become sparse as such archives predominantly depend upon constrained thermal boundaries. For a detailed review of Mediterranean climate variability and change over the last centuries and the availability of existing proxies, the reader is referred to Luterbacher et al. (2006). Dendroclimatological studies from the Central Pyrenees are limited to TRW data from living trees (e.g., Camarero et al. 1998; Gutiérrez 1991; Rolland and Schueller 1994; Ruiz-Flaño 1988; Tardif et al. 2003). Existing studies mainly focus on the estimation of high to mid frequency variations, as missing relict material hinders the preservation of lower frequency variations via age-related composite tree-ring detrending methods (Briffa et al. 1992; Esper et al. 2003). Lower resolution archives, such as chrysophyte cysts from lake sediments collected at >100 lakes from higher elevations in the Central and Eastern Pyrenees, revealed long-term information of winter/spring temperature fluctuations in the northwestern Mediterranean region (Pla and Catalan 2005).
With respect to the past millennium and EU, only a few temperature sensitive composite TRW records (i.e., combining recent and relict wood) exist from northern Scandinavia (Briffa et al. 2007; Helama et al. 2005) and the Alps (Büntgen et al. 2005; Nicolussi and Patzelt 2000), but none from the Mediterranean region. In addition, many of the existing tree-ring sites were sampled in the 1970–1980s, and thus miss the most recent warming trend seen in instrumental station measurements across the greater Alpine region (Auer et al. 2007).
Here we present the first tree-ring dataset combining samples of living and dry-dead timberline wood from the Central Spanish Pyrenees that reaches back prior to AD 1000 and extends forward into the twenty-first century (924–2005). As this study was performed to best preserve inter-annual to multi-centennial scale summer temperature variations, MXD measurements were processed for a selection of 261 cores that meet distribution criteria necessary for an optimized estimation of long-term trends. Age-related tree-ring detrending and variance adjustment methods were applied for chronology development, and various meteorological datasets considered for the analysis of monthly resolved growth/climate responses, calibration exercises, and spatial field correlations. Results allowed the full range of regional high to low frequency May-September maximum temperatures to be estimated for the past eight centuries. This southern EU temperature history was first compared with latest findings from the Alps and Scandinavia and then with prominent large-scale reconstructions, all covering the past millennium.
2 Data and methods
2.1 Tree-ring data and detrending
Characteristics of the three MXD datasets used to reconstruct long-term regional-scale summer temperature variations
2.2 Meteorological data and statistical analysis
Monthly minimum and maximum temperatures from the Pic du Midi mountain observatory (Pic du Midi de Bigorre: 2,862 m asl, 43°04′N, 0°09′E) that start in 1882 were used for proxy calibration. While the temperature data are described to have sufficient quality (Bücher and Dessens 1991; Dessens and Bücher 1995), less value is reported for the precipitation measurements (Dessens and Bücher 1997). A 5° × 5° grid of homogenized and variance adjusted land and sea surface temperature data back to 1850 was used for spatial field correlations (HadCRUT3v; Brohan et al. 2006). For comparison with the Pic du Midi station measurements and the MXD-based estimates, we selected values from a single grid-box centered over 42.5°N and 2.5°E. The HadCRUT3v grid was considered as temperatures of the 1850–2003 period and their homogenization allows longer-term trends to be reasonably well preserved (Brohan et al. 2006). Since precipitation sums were, however, not available from this compilation, monthly temperature means and precipitation sums from a higher resolution (0.5 × 0.5°) grid were utilized for twentieth century (1901–2002) growth/climate response analyses (CRUTS2.1; Mitchell and Jones 2005). Mean values from 42 grid-boxes covering the 42–43°N and 2°W–3°E region, and averages from three longitudinal sub-regions (0–2°W, 0–2°E, 2–3°E) were selected. Due to the shorter period covered, the interpolation techniques performed, and the homogenization applied, potential longer-term trends in this data are not fully preserved (see details in Mitchell and Jones 2005).
To estimate local-scale climatic conditions, particularly those of the GER site, three nearby high-elevation instrumental station records were kindly provided by the Meteorological Institute of Catalonia: BONAIGUA (2,263 m asl, 42°40′N, 1°06′E), SANT MAURICI (1,920 m asl, 42°34′N, 1°00′E), and ESTANY-GENTO (2,120 m asl, 42°30′N, 1°00′E). Mean annual temperature (1961–1990) of the three stations is 4.3°C (2.1 SD). Lowest (−2.5°C; 2.5 SD) and highest (13.1°C; 1.3 SD) monthly values are reported for January and July, respectively. Mean annual temperature averaged over four 30-year reference windows (1931–1960, 1941–1970, 1951–1980 and 1971–2000) ranges from 3.3 to 4.6°C (1.9–2.2 SD). Mean annual precipitation averaged over these periods ranges from ∼1,170 to 1,300 mm (∼58 to 65 SD). The evenly distributed amount of annual precipitation most likely results from the high-elevation location of the instrumental stations (and study sites), which are receiving a constant flow of maritime (Atlantic) air masses all year long. Data were transformed to anomalies with respect to 1961–1990, and significance levels ‘conservatively’ corrected for lag-1 autocorrelation (Trenberth 1984).
A split calibration/verification approach was performed to assess temporal stability of the transfer model, with the following metrics being considered: Pearson’s correlation coefficient (r), reduction of error (RE), and Coefficient of Efficiency (CE). Both RE and CE are measures of shared variance between actual and estimated series (CE is a more rigorous verification statistic), with a positive value suggesting that the reconstruction has some skill (Cook et al. 1994).
3 Results and discussion
3.1 Chronology characteristics
Regional curves and RCS chronologies were separately calculated for TRW and MXD data of the GER and SOB sites. While estimated TRW growth trends are remarkably similar at both sites (Fig. 3d), a slight but systematic level offset between the MXD growth trends is seen. The RC based on the 58 SOB series indicates generally higher density compared to the RC of the 203 GER series, with largest differences found during juvenile growth. Resulting RCS chronologies of the GER and SOB MXD measurements correlate at 0.53 over their common 1517–2005 period of ≥5 series (Fig. 4a), with little differences (0.57 and 0.49) being obtained from their 20-year high- and low-pass fractions (Fig. 4b, c).
To account for the differing MXD growth rates between GER and SOB, two independent RCs were used for RCS detrending on a site-by-site level. The detrended 203 GER and 58 SOB index series were then averaged to create the final, variance adjusted Pyrenees chronology (hereinafter PYR). To avoid potential biases during the record’s period of site overlap, we shifted the 58 SOB series by the average difference between the GER and SOB mean records over 1517–2005. This systematic offset is minimal with a mean difference of 0.0013 index units.
3.2 Growth/climate responses
A similar relationship between radial growth and climate of several Pinus uncinata TRW (near timberline) sites from the Central Spanish Pyrenees has been observed (Tardif et al. 2003). For more details on the growth/climate response of various tree-ring parameters in the Pyrenees, see Büntgen et al. (2007b). A distinct response optimum of MXD to maximum growing season temperatures is found in British Columbia, Canada (Luckman and Wilson 2005; Wilson and Luckman 2003), and also in the Altai Mountains, southern Russia (Frank et al. 2007a). Comparable patterns of MXD formation, i.e., strong correlation with temperature during the early and late vegetation period with weaker correlation in between, are reported from a high-elevation larch network in the Swiss Alps (Büntgen et al. 2006b), from a multi-species network across the greater Alpine region (Frank and Esper 2005a), and from hundreds of sites along the northern latitudinal timberline (Briffa et al. 2002). Some altitudinal/latitudinal modification of the absolute growing season length, however, should be taken into account, when comparing results from such different geographical regions.
3.3 Temperature reconstruction
Common features between the measured and estimated maximum May–September temperatures include relative stable temperatures from ∼1880 to 1950, a decrease from ∼1950 to 1980, and a most recent increase from ∼1980 to present. No anomalous and systematic divergence between (warmer) actual and (cooler) estimated temperatures during the past decades, as reported from some tree-ring sites in the Alps (Büntgen et al. 2006a), and across the northern latitudes (Briffa et al. 1998), for example, is observed. This ability of tracking the most recent summer warmth is in line with revised network analyses from the Alpine arc (Büntgen et al. 2008).
As an additional comparison of the new PYR reconstruction, we utilized the average of the nearest grid-points (0–1.5°E; 41–43°N) of the multi-proxy EU June-August temperature reconstructions (1500–2002) derived from Luterbacher et al. (2004). A significant correlation (at the 99.9% level) was revealed throughout the last 500 years. Higher correlations of the post 1750 period and lower ones before might result from increasing uncertainties in both reconstructions back in time.
Uncertainty in the PYR record is most evident on decadal time-scales. Potential sources of bias include: (1) Relict material from both sampling sites becomes exceptionally scarce in the thirteenth century, causing insufficient chronology replication prior to AD 1260. (2) Variable degrees of relict wood decay can influence the stem coring location, introducing deviations from the standard sampling location at breast height, and thus hinder the estimation of germination ages. (3) The overall long-term ‘shape’ of RCS chronologies is somewhat insecure (i.e., relative level of the recent warmth compared to conditions during the records earliest portion), as various implications of data and methodology are not yet fully quantified (Esper et al. 2003; Helama et al. 2005; Melvin 2004). (4) A limited number of ‘proper’ (i.e., length, homogenization, parameter) instrumental station data that reflect climate conditions of the high-elevation sampling sites, hampers calibration/verification trials to be performed over longer intervals.
3.4 Temporal variability
Commonly reported ups and downs in-phase with the so-called LIA Type Events—a term introduced by Wanner et al. (2000)—most likely refer to modifications in atmospheric circulation patterns during the Wolf (1290–1320), Maunder and Dalton solar minima (Luterbacher et al. 2001, 2002; Wanner et al. 1997; Xoplaki et al. 2001). The later (∼1850) is associated with the most extended Alpine Holocene glaciers advance (e.g., Holzhauser et al. 2005). Interestingly, most Scandinavian glaciers reached their late Holocene maximum during the early to mid 18th century (Nesje and Dahl 2003 and references therein), and glaciers in northern Sweden and Norway, mostly those located on the more continentally influenced eastern slope of the Scandes, advanced until the beginning of the twentieth century (Karlén 1988). Besides such local-scale variation in late Holocene glacier fluctuations across Scandinavia, causes for asynchronous LIA maxima between the Alps and Scandinavia remain unclear. Since rapid fluctuations in mass-balance, particularly of maritime glaciers in western Scandinavia, are driven in part by winter precipitation (Nesje et al. 2000), caution is advised when comparing such archives with tree-ring based summer temperature reconstructions.
Moving 51-year correlations between the three reconstructions show relatively high coherency between the Alps and Pyrenees back to ∼1450, whereas correlations with the Scandinavian data are generally lower and temporally unstable (Fig. 9b). Over the AD 1260–2003 common period, the correlation between the Pyrenees and Alps is 0.34, remains the same after 40-year high-pass filtering, but increases to 0.41 and 0.57 after 40-year low-pass and 20–40 year band-pass filtering, respectively (Fig. 9c). Non-significant correlations are found between Scandinavia and either the Alps or the Pyrenees. The strongest agreement between the Pyrenees and Alps is obtained on the mid-frequency domain, likely due to a combination of the proxy’s unexplained variance in longer-term fluctuations, plus some regional-induced differences in the preserved high-frequency signal. Correlation between the Pyrenees and Alpine proxy data is 0.45 over the 1882–2003 period. Correlation between the instrumental target data from these regions is 0.51 over the same period. Correlation between the Scandinavian reconstruction and records from the Alps and Pyrenees is 0.11 and 0.01, respectively (1882–2003). Interestingly, similarly low correlations of 0.14 and 0.06 are obtained when using the Scandinavian instrumental targets and those from the Alps and Pyrenees, respectively.
3.5 Spatial variability
Overall, the three regional patterns of spatial field correlations as derived from either the proxy or instrumental data are similar. A clear synoptic separation between northern Scandinavia and central EU is emphasized, whereas the greater Alpine region and Mediterranean basin are both influenced by the same synoptic regimes. These results are consistent for various frequency bands and time periods. See Raible et al. (2006) and Xoplaki et al. (2003, 2004) for a detailed description of long-term EU climate variability derived from instrumental observations, proxy reconstructions, and model simulations.
3.6 Large-scale comparison
The Scandinavian and Alpine reconstructions, in-phase with the NH records generally portray high temperatures centered ∼1000 (Fig. 9). Low values during the Oort solar minimum ∼1040 to 1080 are most evident in the Alpine, and partly in the NH records (Fig. 12). Relatively high temperatures in the second half of the twelfth century are common to all reconstructions, although decadal-scale divergence during that early period is evident. While the Scandinavian, Pyrenees and NH reconstructions indicate low temperatures for the second half of the thirteenth century those estimated for the Alpine region are relatively warm. Temperatures during the sixteenth century are estimated to be warm in Scandinavia, cold in the Alps, and about average in the Pyrenees and NH records. A prolonged period of cold summers from ∼1600 to 1850 is most evident in the Pyrenees and Alps, with comparable values seen in the NH records. While warmest summers in the Alps, Pyrenees, and the NH are documented for the twentieth century, Scandinavian summers are reconstructed to be about average during that period. The depression ∼1970 prior to the most recent warming is distinct for the Pyrenees and Alps, but less pronounced in Scandinavia. None of the NH reconstructions allows the most recent decade of warming (e.g., Brohan et al. 2006) to be placed in a longer-term context, as records end between 1979 and 1996. The dominance of TRW data further complicates benchmarking climatic extremes, as annual variations in TRW reflect a shorter portion of the high summer season with a tendency of containing some effects of the previous year climate. In contrast, variations in MXD capture temperatures of an extended season with reduced biological persistence (e.g., Frank and Esper 2005a).
Our collection of living and dry-dead wood from two timberline sites in the Central Spanish Pyrenees results in a composite dataset spanning the AD 924–2005 period. Two chronologies developed on a site-by-site basis correlate at 0.57 over the 1517–2005 common period. The new Pyrenees chronology correlates at 0.53 with maximum May–September temperatures from the Pic du Midi (1944–2005). The common signal is weighted towards higher frequency variations, as correlation increases to 0.68 after 15-year high-pass filtering. The final reconstruction covers the 1260–2005 period and reveals relatively high temperatures in the 14–15th and twentieth century, separated by a rather prolonged cooling from ∼1450 to 1850. The six warmest decades occurred during the twentieth century, with the following four being reconstructed for the period 1360–1440. Comparison with summer temperature reconstructions from the Swiss Alps and northern Scandinavia indicates decadal to longer-term similarity between the Pyrenees and Alps, but notable independence for northern EU. Twentieth century warmth is evident in the Alps and Pyrenees, whereas Scandinavian temperatures are relatively cold compared to those centered ∼1000, 1400 and 1750. Temperature depressions during the second half of the fifteenth century, between ∼1600 to 1700, and ∼1820 are common features of all reconstructions. Lowest temperatures in Scandinavia, however, occurred at the onset of the twentieth century. Spatial field correlations using proxy and target data reveal similar patterns. Both describe a distinct separation between the northern Scandinavian and Alpine spatial fields of correlations, but significant overlap of the Alpine and Pyrenees clusters. Comparison of the three regional-scale studies with six large-scale records suggests that decadal-scale fluctuations of EU summer temperatures are fairly synchronous with those reported from the NH, i.e., common decadal-scale depressions occurred around ∼1350, 1460, 1600, 1700, 1820, and 1970. The Pyrenees MXD record fills a spatial gap in the worldwide tree-ring density network. This gap happens to coincide with lower latitudes, which are generally underrepresented in terms of long-term temperature data. The new reconstruction improves our understanding of past EU temperature variations, and will contribute to the enhancement of NH tree-ring compilations.
The comparison of the three MXD-based summer temperature reconstructions shows divergence in their long-term variations, which allows spatiotemporal patterns in past EU temperatures to be distinguished. These regional discrepancies further indicate the complexity of continental-scale climate variability. Therefore, future research will need to consider (1) the update of existing (e.g., covering the entire Pyrenees from the Mediterranean Sea in the east to the Atlantic Ocean in the west), and (2) development of new regional-scale composite chronologies (e.g., providing tree-ring evidence from the Carpathian arc). Such data should exclusively be derived from (3) high-elevation timberline sites, containing (4) relict and sub-fossil wood, and with (5) MXD measurements being performed.
We thank F.H. Schweingruber for site selection, the National Park d’Aigüestortes I Estany de Sant Maurici (Jordi Vicente i Canillas) for sampling permission and logistic support, R.J.S. Wilson for field assistance and discussion. J. Dessens kindly provided instrumental data from the Pic du Midi. Spatial field correlations were generated using the KNMI Climate Explorer (http://www.climexp.knmi.nl). Supported by the SNF project NCCR-Climate and Euro-Trans (#200021-105663), and the EU project Millennium (#017008-2).
- Auer I, and 31 co-authors (2007) HISTALP—Historical instrumental climatological surface time series of the greater Alpine region 1760–2003. Int J Climatol 27:17–46Google Scholar
- Briffa KR, Shishov VV, Melvin TM, Vaganov EA, Grudd H, Hantemirov RM, Eronen M, Naurzbaev MM (2007) Trends in recent temperature and radial tree growth spanning 2000 years across northwestern Eurasia. Philos Trans Roy Soc B. doi:10.1098/rstb.2007.2199
- Brohan P, Kennedy JJ, Harris I, Tett SFB, Jones PD (2006) Uncertainty estimates in regional and global observed temperature changes: a new dataset from 1850. J Geophys Res 111:D12106. doi:10.1029/2005JD006548
- Büntgen U, Frank DC, Schmidhalter M, Neuwirth B, Seifert M, Esper J (2006a) Growth/climate response shift in a long subalpine spruce chronology. Trees Struct Funct 20:99–110Google Scholar
- Büntgen U, Frank DC, Kaczka RJ, Verstege A, Zwijacz-Kozica T, Esper J (2007a) Growth/climate response of a multi-species tree-ring network in the Western Carpathian Tatra Mountains, Poland and Slovakia. Tree Physiol 27:689–702Google Scholar
- Büntgen U, Frank DC, Verstege A, Nievergelt D, Esper J (2007b) Climatic response of multiple tree-ring parameters from the Central Spanish Pyrenees. Trace 5:60–72Google Scholar
- Büntgen U, Frank DC, Wilson R, Esper J (2008) Testing for tree-ring divergence in the European Alps. Global Change Biol (in press)Google Scholar
- Cook ER (1985) A time series analysis approach to tree-ring standardization. PhD Thesis, University of Arizona, pp 171Google Scholar
- Cook ER, Peters K (1981) The smoothing spline: A new approach to standardizing forest interior tree-ring width series for dendroclimatic studies. Tree-Ring Bull 41:45–53Google Scholar
- D’Arrigo R, Wilson RJS, Jacoby GC (2006) On the long-term context for late 20th century warming. J Geophys Res 111:D03103. doi:10.1029/2005JD006352
- Esper J, Cook ER, Krusic PJ, Peters K, Schweingruber FH (2003) Tests of the RCS method for preserving low-frequency variability in long tree-ring chronologies. Tree-Ring Res 59:81–98Google Scholar
- Esper J, Frank DC, Wilson RJS, Briffa KR (2005a) Effect of scaling and regression on reconstructed temperature amplitude for the past millennium. Geophys Res Lett 31. doi:10.1029/2004GL021236
- Esper J, Frank DC, Büntgen U, Verstege A, Luterbacher J, Xoplaki E (2007) Long-term drought severity variations in Morocco. Geophys Res Lett 34. doi:10.1029/2007GL030844
- Frank DC, Ovchinnikov D, Kirdyanov A, Esper J (2007a) The potential for long-term climatic reconstructions in the Central Altay Mountains from living and relict larch. Trace 5:85–96Google Scholar
- Frank DC, Esper J, Cook E (2007b) Adjustment for proxy number and coherence in a large-scale temperature reconstruction. Geophys Res Lett 34. doi:10.1029/2007GL030571
- Frank DC, Büntgen U, Böhm R, Maugeri M, Esper J (2007c) Warmer early instrumental measurements versus colder reconstructed temperatures: shooting at a moving target. Q Sci Rev. doi:10.1016/j.quascirev.2007.08.002
- Fritts HC (1976) Tree rings and climate. Academic Press, London, pp 567Google Scholar
- Grove JM (1988) The little ice age. Methuen & Co., London, New York, pp 498Google Scholar
- Grudd H (2008) Torneträsk tree-ring width and density AD 500–2004: a test of climatic sensitivity and a new 1500-year reconstruction of northern Fennoscandian summers. Clim Dyn. doi:10.1007/s00382-007-0358-2
- Gutiérrez E (1991) Climatic tree growth relationships for Pinus uncinata Ram. in the Spanish pre-Pyrenees. Act Oecol 12:213–225Google Scholar
- Luterbacher J, 48 co-authors (2006) Mediterranean climate variability over the last centuries: A review. In: Lionello P, Malanotte-Rizzoli P, Boscolo R (eds) The Mediterranean Climate: an overview of the main characteristics, issues. Elsevier, Amsterdam, pp 27–148Google Scholar
- Melvin TM (2004) Historical growth rates and changing climatic sensitivity of boreal conifers. PhD Thesis, Climatic Research Unit, School of Environmental Sciences, University of East Anglia, 271 ppGoogle Scholar
- Osborn TJ, Briffa KR, Jones PD (1997) Adjusting variance for sample-size in tree-ring chronologies and other regional-mean time-series. Dendrochronologia 15:89–99Google Scholar
- Raible CC, Casty C, Luterbacher J, Pauling A, Esper J, Frank DC, Büntgen U, Roesch AC, Tschuck P, Wild M, Vidale PL, Schär C, Wanner H (2006) Climate variability - observations, reconstructions, and model simulations for the Atlantic-European and Alpine region from 1500–2100 AD. Clim Change. doi:10.1007/s10584-006-9061-2
- Rolland C, Schueller F (1994) Relationships between mountain pine and climate in the French Pyrenees (Font-Romeu) studied using the radiodensitometrical method. Pirineos 144:55–70Google Scholar
- Ruiz-Flaño P (1988) Dendroclimatic series of Pinus uncinata R. in the Central Pyrenees and in the Iberian System. A comparative study. Pirineos 132:49–64Google Scholar
- Xoplaki E, Gonzalez-Rouco FJ, Luterbacher J, Wanner H (2003) Mediterranean summer air temperature variability and its connection to the large-scale atmospheric circulation and SSTs. Clim Dyn 20:723–739Google Scholar